DE102018206975A1 - Method for operating a hearing aid and hearing aid - Google Patents

Abstract

A hearing aid (1) according to the invention has an acceleration sensor (6) which, in the intended wearing state, is positioned on the head (9) of a hearing aid wearer and which is set up for measurement in three measuring axes (x, y, z) perpendicular to one another. According to the method, a movement plane (54) of the movement of the hearing aid wearer is derived from the acceleration data (A) from the acceleration data (A) of the acceleration sensor (6), a motion axis (A) is derived from the acceleration data (A) on the basis of acceleration data (A) of the acceleration sensor (6) transported by an acceleration signal. 52) and a movement direction of the movement, and based on the movement plane (54), the movement axis (52) and the movement direction to a presence of a rotational movement of the head (9) closed. From the detected rotational movements, in particular on the basis of a yaw angle (G) determined thereby, a line of sight probability distribution is also created which indicates a probability as to whether the actual viewing direction of the hearing device wearer was along an associated angle.

Description

The invention relates to a method for operating a hearing device and a hearing device, which is set up in particular for carrying out the method.

Hearing aids serve in particular in the form of hearing aids to persons with a hearing loss to compensate for the hearing loss at least partially. For this purpose, conventional hearing aids regularly comprise at least one microphone for detecting ambient noise and a signal processing processor which serves to process the detected sounds and in particular to amplify and / or attenuate them as a function of the individual hearing loss (in particular frequency-specific) , The processed microphone signals are forwarded by the signal processing processor to an output transducer - usually in the form of a loudspeaker - for output to the hearing of the respective hearing device wearer. Depending on the type of hearing loss, so-called bone conduction earphones or cochlear implants for mechanical or electrical stimulation of the hearing are used as output transducers. The term hearing aid but other devices such as headphones, so-called tinnitus maskers or headsets are summarized.

In particular, hearing aids often have a so-called classifier, which serves to close in particular on the basis of the detected sounds on certain predefined "listening situations". Depending on the detected hearing situation, the signal processing is then changed regularly. Since speech understanding of the hearing device wearer is often impaired due to the present hearing loss, the (signal processing) algorithms stored in the signal processing processor are usually tuned to work out the speech utterances of third parties in the recorded microphone signals and reproduce them in the most understandable form possible for the respective hearing aid wearer. To recognize a conversation situation, a speech recognition algorithm is often executed in the classifier. However, such an algorithm becomes inaccurate in situations where several people are speaking in the vicinity of the hearing aid wearer, but not all are participating in the same conversation. In this case, an acoustic identification of the persons participating in the same conversation is regularly difficult.

The invention has for its object to enable improved operation of a hearing aid.

This object is achieved by a method having the features of claim 1. Furthermore, this object is achieved by a hearing aid with the features of claim 26. Advantageous and partly inventive embodiments and refinements of the invention are set forth in the dependent claims and the following Description set forth.

The inventive method is used to operate a hearing aid, in particular a hearing aid. The hearing aid in this case has an acceleration sensor, which is positioned in the intended wearing state on the head, in particular in or on an ear of a hearing aid wearer. This acceleration sensor is also set up for measurement in three mutually perpendicular measuring axes, of which preferably at least roughly in the intended state of wear (ie not necessarily exact, especially - mostly due to the individual anatomy - approximately, ie with a deviation of up to about 20 or 30 degrees) are arranged in a transverse plane of the body of the hearing aid wearer. This means, in particular, that these two measuring axes are assigned to the transversal plane.

According to the method (in particular during normal operation of the hearing aid) based on transported by an acceleration signal acceleration data of the acceleration sensor on a movement of the hearing aid wearer, ie in particular whether the hearing aid wearer moves at all closed. From the acceleration data is also - especially if a movement was detected - derived a movement plane of the movement of the hearing aid wearer. Furthermore, from the acceleration data, a movement axis and a movement direction of the movement are determined, as well as closed on the basis of the movement plane, the movement axis and the direction of movement on a presence of a rotational movement of the head.

The term "transverse plane" is understood here and below as meaning, in particular, a plane oriented perpendicular to (body) longitudinal axis of the hearing device wearer. According to this term borrowed from medicine, the term "sagittal plane" or "median plane" is used here and below, and thus in particular a - in an upright posture and straight-ahead view - the body of the hearing aid wearer in a left and right part, in the case the median plane into a corresponding half, and divide perpendicular to the transverse plane.

The term rotational movement is understood here and below to mean, in particular, a rotation of the head about a (body) axis. In principle, such rotational movements of the head are within the scope of the present invention Description in particular divided into a yaw, a pitching motion and a rolling motion. The term "yawing" or "yawing" is here and below in particular a rotation (rotation) of the head about a (body) vertical axis, ie in particular about the axis formed by the spine (which preferably at least approximately coincides with the vertical ) Understood. Furthermore, here and below, the terms "pitching" or "pitching motion" are used in particular for a movement directed up and down about a "pitch axis" which preferably lies at least roughly in the transverse plane and in particular connecting the ears of the hearing device wearer, and also "rollers" or " Rolling movement "for a sideways tilting or tilting of the head about a preferably in the neutral direction of view (also referred to as" zero-degree viewing direction ") aligned" roll axis ", preferably a section line of the transverse and the sagittal plane at about the ears forms, used. The roll axis is therefore also at least roughly in the transverse plane.

Optionally, the direction of movement is determined only indirectly, for example by means of a rotation angle, for example a yaw angle, which reproduces the extent of the rotational movement (for example the yaw movement) about the respective movement axis and also recognizes a direction of rotation (and thus the direction of movement) via its assigned sign leaves.

By first "only" determining whether there is any movement of the hearing device wearer, optional subsequent process steps (for example for detecting the rotational movement) can be dispensed with. Furthermore, the determination of the plane of motion as well as the axis of movement and the direction of movement advantageously makes it possible to use only one (single) acceleration sensor, which measures in (or: along) three measuring axes, to detect the rotational movement, so that the use of conventionally used measuring systems, the multiple (especially different) sensors, specifically a combination of acceleration sensors with gyroscopes and / or magnetic field sensors (also referred to as "inertial measurement units") use, and the associated relatively high energy consumption can be omitted. Furthermore, the detected rotational motion may be used to assist in the analysis of hearing situations or the adjustment of signal processing parameters.

Preferably, the third measuring axis of the acceleration sensor in the intended carrying position is aligned approximately perpendicular to the transverse plane. Preferably, the acceleration sensor used is an acceleration sensor in which the sensitive elements associated with the respective measuring axis are integrated. This is advantageous in that such "3D acceleration sensors" are available as integral components.

In a further preferred variant of the method, the acceleration data are examined in particular for the presence of the yaw movement as a rotational movement. This often reflects a "head turning", for example, in the context of a conversation situation again, so that a relevant information can be advantageously used for setting signal processing parameters, in particular for a conversation situation with multiple participants. Alternatively or additionally, the acceleration data are examined in particular for the presence of pitching and / or rolling motion.

In a further expedient variant of the method, the acceleration data are analyzed block by block in successive-possibly overlapping-data frames (that is to say, in particular, a plurality of data samples which are acquired over a predetermined time window). The length of the respective data frame (i.e., the allocated time window) is about 0.5 to 1.5 seconds. Preferably, an overlap of the subsequent time window with the previous time window of about 0.25-0.75 seconds is applied. The length of the (respective) time window results from the recognition that a conventional, conscious rotational movement (eg the yaw movement) of the head lasts about 0.5 seconds to about 1.5 seconds. In particular, in this method variant of the acceleration sensor with a frequency of about 10-60 hertz, preferably from about 15-20 hertz in each case two or three measured values, which are assigned to the two or three measuring axes output. These measured value groups, referred to as data samples, are buffered in particular in a buffer memory which, for example, can hold eight of these data samples (which then form a data frame). A so-called "update rate" of the buffer memory is preferably about two hertz.

In a preferred variant of the method, the presence of the movement of the hearing device wearer is in particular closed when a predetermined criterion is fulfilled. In particular, this criterion must be met by an assigned feature (also referred to as "feature") that has been specifically extracted from the acceleration data. In the present case, as such a feature preferably a (in particular measured, ie detected) range of values and / or a variance of the acceleration signal, preferably the acceleration data contained in the respective analyzed data frame is used. Expediently, in this case, the presence of the movement (and thus fulfillment of the criterion) is concluded if the characteristic of the feature (that is to say in particular the size of the value range or the value of the variance) is different from the respectively associated characteristic indicative of measurement noise , For example, this is done for a threshold comparison. This method variant is based on the assumption that the acceleration data in the absence of movement of the hearing aid wearer in particular contain only the value of the gravitational acceleration, systematic measurement errors and measurement noise. In particular, only the measurement noise within a data frame varies with high probability, so that in particular the variance reflects only the change of the measurement noise. In this case, a threshold comparison is a comparatively simple variant for checking the criterion for fulfillment by the corresponding feature.

In particular, for a preprocessing of the acceleration data in the intended wearing state of the hearing aid from the acceleration data for a rotation of the (three) measuring axes of the acceleration sensor relative to the transverse plane and / or the horizontal plane indicative rotation operator (hereinafter referred to as "normalization rotation operator") determined. The normalization rotation operator thus contains information about which axis and by how many degrees the measurement coordinate system of the acceleration sensor has to be rotated in order to be brought into coincidence preferably with the two measurement axes assigned to the transversal plane. Thus, this normalization rotation operator reproduces a "fixed" linkage of the measurement coordinate system with the body coordinate system (in particular in neutral body or at least head posture).

As a rotation operator, here and below, for example, a rotation matrix, Euler angle or a comparable operator are used. Particularly preferred here and in the following as a rotation operator but (in each case) a quaternion (referred to in the method variant described above as "normalization quaternion") is used. To construct the corresponding quaternion, a so-called quaternion-axis-angle formula is used in particular. The use of quaternions for the (description or parameterization of) rotation of bodies or data is advantageous in particular with regard to the computational effort and the stability during the required arithmetic operations.

In the present case, the rotation of the measuring axes is preferably determined on the basis of the cross product and the scalar product between the gravitational vector determined from the acceleration data and a "global" gravitational vector. The global (or even "ideal") gravitational vector is assigned to a "world coordinate system", which is assumed to be universally valid. In this, the global gravitational vector is pointing "upwards", i. H. in particular vertically and oriented outwards from the center of the earth (g = [0, 0, 1]), in particular since the acceleration sensor regularly outputs the gravitational acceleration in such a way. To determine the twist, it is further assumed that when the head (and especially the entire body of the hearing aid wearer) is in the neutral ("upright") position, the head and world coordinate systems are aligned the same. In this case, the transverse plane of the body at least approximately coincides with the horizontal plane. By means of the cross product is thereby a tilting axis, around which the "rotation" in the correction (or: normalization) must be determined, and determined on the basis of the scalar product, the angle by which the rotation about the tilt axis for the correction must be made. Ie. between the measured value of the gravitation in the sensor coordinate system and the ideal gravitational vector in the world coordinate system, the cross and scalar product is calculated and from this the normalization rotation operator, in particular the normalization quaternion, is determined.

In a convenient method variant, the acceleration data is smoothed, i. any signal peaks present are removed, in particular by using a median filter. Its application, in particular its parameterization depends in particular on the length of the data frame and is set to, for example, three data samples. Other filters are also conceivable.

In a preferred variant of the method, an acceleration directed radially toward the head of the hearing aid wearer (hereinafter referred to as "radial acceleration") and a tangentially directed acceleration (hereinafter also referred to as "tangential acceleration" for short) are derived from the acceleration data for the corresponding rotational movement , For the (expedient) case that one of the transversal plane associated measuring axes in the proper wearing state is also - at least approximately - aligned along a tangent of the head, the acceleration data associated with this measuring axis - in a yaw movement of the head - expediently a high proportion of tangential Acceleration. In this case, the second measuring axis assigned to the transversal plane shows in particular a high proportion of the radial acceleration. Therefore, in a useful sub-variant as indicative of the yawing movement a time course of considered and evaluated by means of the transversal level associated measuring axes determined tangential and radial acceleration. As a criterion for the presence of the yaw motion is used in this case and considered whether the time course of the tangential acceleration within a given movement time window consecutively two oppositely directed local extremes (ie, for example, a local maximum and a local minimum). In particular, it is considered whether, over time, the tangential acceleration for these two extrema assumes values with opposite signs. This is based on the knowledge that when yawing the head, the tangential acceleration initially indicates an "actual" acceleration and then a "negative" acceleration (namely when braking the head) with a respective associated rash (the respective extremum) over time. In particular, depending on the orientation of the measuring axis associated with the tangential direction relative to the actual direction of movement, the tangential acceleration thus assumes, for example, first positive values and "changes" when the head is decelerated to negative values. When yawing the head in the opposite direction, the values of the tangential acceleration change accordingly from negative to positive. In this variant of the method, the movement time window is preferably adapted to the duration of a head-turning movement usual in a group discussion, and preferably has values between 0.25 and 1.5 seconds, in particular from 0.5 to 1 second. The movement time window is preferably "opened" (ie its monitoring is started) if a sufficiently significant change in the values of the tangential acceleration is detected. The movement time window advantageously achieves a (temporal) limitation of a consideration of the main feature, in particular the temporal course of the tangential acceleration, so that "acceleration events" which, due to their comparatively long duration, are very likely to be associated with no head rotation (thus no yaw) , not considered. The radial acceleration is also known to be associated with the centripetal force (or the counter-rotating centrifugal force) inevitably occurring during a rotational movement (in particular the yawing motion), so that-with a temporal evaluation of the radial acceleration-it is expediently investigated as a further feature whether within the Movement window is a local extremum of the radial acceleration. In connection with the two local extremes of the tangential acceleration, it is advantageously possible to derive a comparatively high probability that not only is there a linear movement along one of the measuring axes, but actually a rotational movement; in the case of the measuring axes associated with the transverse plane, the yawing motion of the head.

The above-described analysis of the acceleration data with respect to the yawing motion can also be applied correspondingly to the rolling and / or pitching motion. In each case measuring axes are evaluated, which can be used for the respective movement plane when rolling or pitching as radial or tangential acceleration axes. Thus, the measuring axes to be evaluated when rolling, in particular by the arranged approximately in a frontal plane of the body measuring axes, and when nodding arranged approximately in the sagittal plane measuring axes. In the latter case, it is advantageous to arrange the acceleration sensor on the hearing aid in such a way that the measurement axis assigned to the vertical (in particular the approximately vertically oriented measurement axis) does not intersect with the pitch axis of the head in the intended wearing state. In this case, this measurement axis is aligned to a tangent of the head in pitch direction parallels. In both cases - d. H. in the evaluation with respect to the rolling or pitching movement - the influence of the gravitational acceleration is expediently compensated dynamically in particular, since this changes constantly depending on the rotation angle of the head. For this purpose, the relationship between the sensor coordinate system and the world coordinate system determined by means of the normalization rotation operator is optionally used. Furthermore, a binaural difference of the acceleration data of both sensors is used, in particular when using a binaural hearing aid system with two substantially identical hearing aids and thus also two acceleration sensors, in order to compensate for the influence of the gravitational acceleration.

In a preferred embodiment, the plane of motion is constructed from the radial acceleration and the tangential acceleration, in particular by these being plotted against one another in a diagram. In doing so, the curve of the radial and the tangential acceleration presented here concretely spans the plane of motion. The plane of motion (and thus also the entire acceleration data contained in the acceleration signal) is then rotated, in particular using a rotation operator, preferably a quaternion, in such a way that it is parallel to one of the two for the detection of the radial acceleration and the tangential. Acceleration used measuring axes clamped measuring plane is aligned. As a result, the radial and the tangential acceleration are linked to the two measurement axes spanning the measurement plane, in particular aligned correspondingly parallel thereto. In the event that that Existence of the yaw motion is investigated, as described above, in the case of the measuring axes used for detecting the radial acceleration and the tangential acceleration, in particular the measuring axes assigned to the transversal plane.

Since the acceleration sensor is set up for measurement in the three measuring axes which are perpendicular to one another, the acceleration data and therefore in particular also the plane of motion for the yawing motion are shifted by a corresponding amount from the measuring plane assigned to the transversal plane due to gravity. Preferably, therefore, after the "correct rotation" of the plane of movement (also: "yaw plane") parallel to the transversal plane, a projection of this into the measuring plane takes place. In particular, this method variant converts the acceleration data from 3D to 2D. Due to the rotation of the plane of motion, the acceleration data are preferably lossless, at least virtually loss-free for further evaluation in the measurement plane described above, so that a particularly precise subsequent evaluation based on the acceleration data, in particular the movement plane derived therefrom, is made possible.

In an expedient development for the rotation of the plane of motion, the tilt of the plane of movement relative to the (above described) measuring plane determined by fitting a curve, preferably an ellipse in a course of radial and tangential acceleration against each other (ie in the course or the Curve fitted). For the surface of the fitted ellipse ("ellipse surface"), a normal vector is subsequently determined and this, in particular the normal vector arranged in the center of gravity of the ellipse surface, is assumed as a movement axis (thus in particular as a yaw, roll or pitch axis). For this movement axis (that is to say for the normal vector), a tilt angle with respect to the measurement plane (in particular with respect to the normal vector of the measurement plane, preferably with the measurement coordinate system oriented to the transverse plane to the normal vector of the transverse plane or to the (possibly global) gravitational vector) is determined. Particularly preferably, a correction rotation operator, in particular a correction quaternion, is constructed from the thus determined tilting of the movement axis, by means of which the movement plane can be turned onto the measurement plane in a simple manner.

In addition, by means of the movement plane, in particular by means of the fitted ellipse and the movement axis, it is particularly preferable to construct a motion rotation operator, in particular a motion quaternion, for the current movement of the head of the hearing device wearer. For this purpose, in particular for the detected rotational movement, a rotation angle is determined from the associated tangential acceleration. In particular, a yaw angle-which indicates the rotation of the head about its vertical axis (i.e., the motion axis described above) -is determined from the tangential acceleration (which in turn is preferably determined from one of the measurement axes associated with the transverse plane). Optionally, as a rotation angle (in addition to or as an alternative to the yaw angle), a roll and / or a pitch angle is determined from the respective associated tangential acceleration. In particular, the corresponding tangential acceleration is integrated twice, for example by means of a high-order Runge-Kutta method. The result of the integration is then preferably divided by the head radius (optionally measured during an initial adaptation of the hearing device or deposited as, for example, a median of a database) to obtain the "new" or current angular position of the head on the "virtual" unit circle. Expediently, during a (first) adaptation of the hearing device to the hearing aid wearer, a rotational radius is also determined for each of the rotational movements described above, with which the acceleration sensor is arranged relative to the respective axis of rotation, ie the yaw axis, the roll axis or the pitch axis. This radius of rotation corresponds in particular in the case of the yawing and rolling movement regularly the head radius for this rotational movement and is optionally used to determine the angular position. In the case of the pitching movement, this radius of rotation is usually smaller than the actual head radius, since the acceleration sensor, in particular the hearing device containing it, is regularly worn on or in the ear and is thus positioned comparatively close to the pitch axis.

Alternatively, in the context of the invention, the above-described evaluation of the movement, in particular the checking for the presence of the rotational movement, preferably the yawing motion and the determination of the new angular position by other estimation methods, for example. By regression methods such as a non-linear autoregression network with exogenous entries ( "NARX"), a recurrent neural network ("RNN") or a "standard least squares" regression method. Feature-based classification methods are also conceivable within the scope of the invention.

In order to save computing capacity during the operation of the hearing device, in an expedient development the rotation angle for the rotational movement of the head (in particular the yaw angle for the yaw movement) is determined only although actually the presence of the corresponding rotational movement (ie, for example, the yaw movement) has been detected.

Appropriately, as (optionally further) feature for the presence of the rotational movement, in particular the yaw, the rolling and / or pitching movement of the head of the hearing aid wearer in particular a surface (in particular the surface area) of the ellipse described above determined and depending on the expression of the Feature (ie, the value of the area) is closed to the presence of the rotational movement. The surface area of the ellipse is indicative of the strength of the rotational movement. In particular, the strength of the rotational movement also increases with increasing surface area, so that, for example, a threshold value comparison for detecting the rotational movement of the head can be carried out. For example, a rotational movement (in particular, a pure yaw movement) of the head over the associated strength is offset by a movement of the entire body (eg, getting up, running, sitting down, etc.) or other head movements, such as head movement. As a head sinking while eating, since this usually lower values for the tangential acceleration are detected. Thus, for example, a distinction of a yaw (especially only) of the head of other body rotations is possible.

As an alternative or additional feature for the presence of the rotational movement (eg of the yawing movement) of the head of the hearing device wearer, in a further expedient process variant, an associated intensity (or also "strength") and / or an intensity of a tangential acceleration is calculated from the tangential acceleration. Derived speed. In particular, the respective intensity is determined by determining the norm of the respective vector or the respective quaternion assigned to the acceleration or the velocity, for example by means of scalar multiplication with itself. If the tangential acceleration or the tangential velocity has a predetermined value, this becomes, in particular, analogous to the above evaluation of the area of the fitted ellipse, in particular as a criterion for the presence of the rotational movement and / or also for differentiation with respect to movement of the entire body the hearing aid wearer used.

In order to reduce and / or avoid systematic errors, in particular in the determination of a yawing motion, preferably a current yaw position of the head, in an advantageous variant of the method, in particular in the event that no movement of the hearing aid wearer is detected, a correction, in particular first of all an identification a pitch angle and / or a roll angle of the head performed. This correction or identification is thus preferably carried out in so-called static data frames. Systematic influences (a so-called bias), which are frequently subject to temperature-induced changes, are known to change extremely slowly compared to the length of the respective data frame (of the order of about 1 second), so that these are used for a data frame as a constant measured value shift ("offset ") Can be accepted. The acceleration data associated with the static data frames thus contain, apart from this offset, in particular only the influence of the gravitation and an additional measurement noise (for example, represented by white Gaussian noise) with a mean value of zero.

Preferably, an average value is initially formed within the respective static data frame for the acceleration data, as a result of which only the bias or offset described above is contained in the acceleration data and thus identified. This bias can now be removed by subtraction, in particular in subsequent data frames.

Furthermore, it is expedient (before or after the averaging described above) to carry out a correction of the acceleration data for the influence of gravity. In particular, for this purpose, the influence of the gravitational acceleration on the respective measuring axes - which is known with known alignment of the measuring axes to the world coordinate system - is subtracted.

In a preferred embodiment of the method variant described above for correcting or identifying the pitch or roll angle, the acceleration data, in particular the associated mean value, preferably its associated vector of the current (static) data frame by means of the alignment information of the preceding data frame (for example by means of this for rotated correction operator or the correction quaternion, or another, as for the alignment of the head in the previous data frame representative set rotational operator) rotates, in particular in order to align the acceleration data parallel to the measuring or- transverse plane can. Subsequently, for the thus rotated (also "estimated") acceleration data preferably by means of the cross product and the scalar product a residual deviation of the alignment compared to the previous data frame, in particular to the gravitational vector of the previous data frame (or alternatively to the global gravitational vector) and thus indirectly to real alignment of the measuring axes determined. Concrete therefore become comparable to the previous description, a correction axis and a correction angle determined.

By means of this correction axis and the correction angle (that is to say by means of the residual deviation determined), it is preferable to construct a deviation rotation operator, in particular a deviation quaternion, which tilts the acceleration data with respect to the preceding data frame and thus also the deviation with respect to pitch and roll angles relative to the previous data frame. Since there should be no movement in a static data frame, the respective deviation with respect to the pitch and roll angles is used in particular as a correction variable.

In a further expedient variant of the method, in particular for the case in which no movement of the hearing aid wearer is detected, a correction of the yaw angle is carried out. For this purpose, a main viewing direction, which conventionally lies in a sagittal, concretely in the median plane of the hearing aid wearer, is adapted from this sagittal or median plane of the body of the hearing aid wearer to a current viewing direction (also: drifted). In other words, the current value of the determined yaw angle is drifted towards zero. As a result, the current viewing direction is adapted as the main viewing direction, in particular over a period of at least one minute, in particular over several minutes. This is especially in lecture or meeting situations (especially with a presentation) advantage in which the hearing aid wearer over a comparatively long period (at least one minute, especially several minutes) obliquely to the sagittal plane to the other interlocutors (with comparatively little variation of his Head posture) or looking at a screen and only occasionally, and in particular briefly turn his head to other participants. In addition, this may also possible errors in an "initialization" in particular a rotation operator, in particular a Quaternion that is indicative of the orientation of the head of the hearing aid wearer (and for the or preferably at initialization a yaw angle of zero degrees as the main direction of vision is accepted, but which does not necessarily coincide with the median level), be compensated ("post-corrected"). Such errors can occur, for example, due to the absence of an absolutely measuring position sensor, for example a magnetometer.

For the above correction of the yaw angle, a rotation operator designated as "yaw-free rotation operator" is preferably determined. In the event that the orientations and rotations of the head or the hearing aid are described in space by means of quaternions, in particular a "greedy quaternion" is determined. Specifically, for this purpose, the yaw angle is set to zero degrees at the above-described deviation rotational operator constructed by the correction axis and the correction angle for the correction of the pitch angle (in particular, in the deviation quaternion). In particular, this yaw-free rotation operator is thus determined on the basis of the (optionally global) gravitational vector. Subsequently, between this yaw-free rotation operator and the deviation rotation operator (in which the yaw angle is not changed) is interpolated spherically and preferably set the resulting rotation operator as a new deviation rotation operator. The step size of the interpolation is chosen, for example, such that the acoustic scene (also: hearing situation) can still be adapted sufficiently well, for example in such a way that the interpolation leads to a comparatively slow drift of the yaw angle. At a current yaw angle of, for example, 45 degrees for a slow drift to zero degrees, the step size z , For example, set to one percent. To 100 Interpolation steps would be the yaw angle - with unmoved head - drifted to zero degrees. With a block length (or length of the respective data frame) of one second, such a complete drift thus corresponds to a time of 100 seconds. Optionally, the ("adaptive") percentage for the interpolation (as well as an optional minimum step size by which the yaw angle is to be drifted) is dependent on (in particular proportional to) a magnitude of an error of the plane estimate (in particular the determination of the motion plane or the fitting of the ellipse in the course of the radial and tangential acceleration, described by a residual error determined in the "Fit"). Furthermore, the percentage is also optionally varied depending on a frequency of head yawing during the last few minutes (for example, the last 2 to 4 minutes). In this case, it is assumed, for example, in the case of comparatively many detected yawing movements, that correspondingly many (accumulated) errors are present and therefore a higher percentage (eg two percent) is used for the compensation. Additionally or alternatively, a maximum value range for the yaw angle of 180 degrees (ie in particular of +/- 90 degrees) is preferably given in the interpolation, since for a pure head rotation (ie without additional rotation of the upper body) no (at least not significantly) larger rotations of the head are possible.

In a preferred method variant, a new orientation rotation operator (in particular an orientation quaternion) is also determined for determining the current orientation of the head, in particular for each analyzed data frame. For this, the preceding (ie the associated with the preceding data frame) with the above-described motion rotation operator (which was designed for a detected motion, particularly a detected yaw motion) or the above-described deviation rotation operator (correspondingly determined in a static data frame). In particular, when using quaternions as a rotation operator, the preceding orientation quaternion is combined with the motion quaternion by means of quaternion multiplication. By means of this (new) orientation rotation operator, for example, the most error-free determination and subtraction of gravitation is possible, because when only one acceleration sensor is used, it is advantageous to know the actual spatial position in order to be able to determine the gravity precisely (in particular without the influence of other accelerations) ,

As is apparent from the above description, as the respective rotation operator, that is to say in particular as normalization, correction, deviation, movement and / or orientation rotation operator, preferably the corresponding quaternion, that is in particular the normalization, correction, deviation, motion or orientation quaternion created.

Particularly preferred for initialization of the orientation rotation operator, ie in particular as a first orientation rotation operator an upright and "neutral" head posture, ie in particular a coincidence of the zero-degree viewing direction with the median plane (and in particular straight, ie preferably along the transverse plane directed view ) accepted. In each case, a value of zero is set for the respective yaw, pitch and roll angles. In the case of the orientation quaternion, the unit quaternion is used in particular. Thus, it is assumed here that a yaw angle of zero degrees represents the zero-degree line of sight (or main sight direction), but the yaw angle does not necessarily have to coincide with the median plane (depending on the actual orientation of the head). In particular, in combination with the above-described correction of the yaw angle (in particular the drift), thus errors between the actual orientation of the head and the median plane can be avoided or at least (post-corrected as described above). An exact alignment of the zero-degree viewing direction to the median plane can thus advantageously be omitted.

The orientation rotation operator of the preceding data frame is preferably used in the context of the above-described correction or identification of the pitch or yaw angle for rotating the mean value, in particular the mean value vector of the subsequent static data frame (ie as preceding orientation information).

In an expedient method variant, a (in particular spatial) line of sight probability distribution is created on the basis of the (in particular respective new) orientation rotation operator (preferably the orientation quaternion). In particular, for each angular position (or "rotational position", in the case of the yawing motion also "yaw position") of the head (in particular starting from the main viewing direction, which is associated with the zero-degree position in particular), this indicates a probability as to whether the actual Viewing direction of the hearing aid wearer within the past time (for example, during the last 1 to 5 minutes) along an angle associated with the respective angular position. Particularly in the case of a conversation situation with several call partners, this line of sight probability distribution will have several, different angular positions, in particular yaw positions, associated rashes ("peaks"), since the hearing aid wearer will in such a situation turn his head to the various call partners. Thus, the line of sight probability distribution also reflects a spatial area of interest in which the conversation partner or other sound source of high interest is for the hearing device wearer with high probability.

In order to establish the line of sight probability distribution, in a preferred development, first of all, starting from the main direction of sight, a Gaussian normal distribution having a predetermined standard deviation is applied. Values of about 5 to 25 degrees, in particular 10 to 15 degrees, are used for the standard deviation. For the current angular position (in particular yaw position) predetermined in particular on the basis of the rotational angle (in particular yaw angle) determined in the above manner, a new normal distribution with predetermined standard deviation (in particular again with the aforementioned values) is then set. Optionally, in particular before setting the new normal distribution, all previously assumed probability values for the angular positions which have occurred so far are uniformly reduced by a predetermined value. Preferably, the newly set normal distribution is also weighted by a factor corresponding to the reduction of the previous probability values, so that the cumulative sum of all angular positions plotted in the line of sight probability distribution yields one. Through this step, in particular, a (preferably increasing) "forgetting" becomes more prevalent (and in particular "old") angular positions achieved by the probability for the presence of an angular position in the absence or sufficient repetition "out" of the line of sight probability distribution. This advantageously leads to a dynamic (in particular automatic) adaptation to the actual situation in which the hearing device wearer is currently located. The advantage of the above-described determination of the line of sight probability distribution lies in particular in the fact that not only a "concrete" value of the current angular position (in particular the current yaw angle) is used, but in each case a band of several angle values which are around the determined (or: estimated) Value of the current rotation angle (in particular the current yaw angle) are distributed around. As a result, errors in the above-described determination of the current value of the rotation angle have a comparatively small effect, in particular because they are "smeared" over a larger angular range. Furthermore, the hearing aid wearer is not likely to assume exactly the same angular position (in particular yaw position) every time he looks at a conversation partner, since he or the conversation partner is moving, and the particular acoustic area of interest usually does not run exactly along the line of sight. Thus, the line of sight probability distribution is comparatively robust against errors.

In an expedient development, the value of the standard deviation for the Gaussian normal distribution is specified variably. For example, the value of the standard deviation is adapted to a possible error in the determination of the yaw angle. Conveniently, the standard deviation (their value) is adjusted concurrently, so that the value is set small for small errors and vice versa. As a result, relatively little weight is assigned to comparatively large errors, and the statement of the line of sight probability (at least for this angular position) is in particular "spongier". As a measure of the error, for example, a quality of the plane fit, in particular the fit of the ellipse on the acceleration data is used. This quality is mapped in particular by a residual error of the plane or ellipticals.

In an expedient variant of the method, the information about the rotational movement, in particular the yawing movement of the head of the hearing device wearer, for example the current yaw position, in particular the line of sight probability distribution, is used to adapt a signal processing algorithm for a group talk situation. In particular, it is determined on the basis of the line of sight probability distribution, in which value range of the respective rotation angle (in particular the yaw angle) were the most frequent views (with Kopfgierung) of the hearing aid wearer, and then, for example, an opening portion of a directional cone adjusted accordingly (eg., Widened or tapered). Optionally, the information about the rotational movement is used in addition to an acoustic classifier, by means of which the existence of a conversation situation with several participants in the conversation is determined per se, preferably in order to verify and / or improve the result. For example, by means of the acoustic classifier in a situation with several conversations between adjacent persons, it is often not possible to ascertain with sufficient certainty which part of speech acquired by means of a microphone system of the hearing device belongs to the conversation.

For example, it can not be ascertained with sufficient certainty in a restaurant whether the hearing aid wearer is talking to the person opposite him or her right or whether the persons to the left of the hearing aid wearer are also participants in this conversation. On the basis of the line of sight probability distribution, it can advantageously be determined whether the hearing aid wearer turns his head to all potential conversation participants or only to a part of them. In the latter case, the chances are high that the other people will not participate in this conversation.

The hearing aid according to the invention has the above-described acceleration sensor as well as a processor which is set up for carrying out the method described above, in particular automatically.

The processor is optionally designed as a non-programmable electronic circuit. Alternatively, the processor is formed by a microcontroller, in which the functionality for carrying out the operating method according to the invention is implemented in the form of a software module.

The conjunction "and / or" is to be understood here and below in particular in such a way that the features linked by means of this conjunction can be formed both together and as alternatives to one another.

• 1 in einem schematischen Schaltbild ein Hörgerät,
• 2 in einer schematischen Aufsicht von oben einen Kopf eines Hörgeräteträgers mit dem bestimmungsgemäß am Ohr getragenen Hörgerät,
• 3 in einem schematischen Ablaufdiagramm ein Verfahren zum Betrieb des Hörgeräts,
• 4 in drei übereinander dargestellten schematischen Diagrammen, jeweils einen Verlauf einer, einer von drei senkrecht aufeinander stehenden Messachsen eines Beschleunigungssensors des Hörgeräts zugeordneten Beschleunigung über der Zeit,
• 5 in einem schematischen dreidimensionalen Diagramm die den drei Messachsen zugeordneten Beschleunigungen gegeneinander angetragen,
• 6 in einer schematischen Darstellung einen Verfahrensschritt zur Ermittlung eines Korrekturwerts für die mittels des Beschleunigungssensors erfassten Beschleunigungsdaten, und
• 7 in einem schematischen Polardiagramm eine Wahrscheinlichkeitsverteilung für eingenommene Blickrichtungen des Hörgeräteträgers.

Embodiments of the invention will be explained in more detail with reference to a drawing. Show:

• 1 in a schematic circuit diagram a hearing aid,
• 2 in a schematic plan view from above a head of a hearing aid wearer with the intended worn on the ear hearing aid,
• 3 in a schematic flow diagram a method for operating the hearing device,
• 4 in three superimposed schematic diagrams, in each case a course of, one of three mutually perpendicular measuring axes of an acceleration sensor of the hearing aid associated acceleration over time,
• 5 in a schematic three-dimensional diagram, the accelerations assigned to the three measuring axes are plotted against each other,
• 6 in a schematic representation of a method step for determining a correction value for the acceleration data detected by the acceleration sensor, and
• 7 in a schematic polar diagram, a probability distribution for assumed directions of the hearing aid wearer.

Corresponding parts and sizes are always provided with the same reference numerals in all figures.

In 1 is a hearing aid 1 , Concretely a so-called behind-the-ear hearing aid shown. The hearing aid 1 includes a (hearing aid) housing 2 in which several electronic components are arranged. As an electronic components, the hearing aid includes 1 two microphones 3 for the detection of noise from the environment of the hearing aid 1 are set up. Furthermore, the hearing aid includes 1 as electronic component a signal processor 4 which is set up by means of microphones 3 to record recorded sounds and output to the hearing of a hearing aid wearer to a loudspeaker 5 issue. To record the spatial position of the hearing aid 1 this also includes an acceleration sensor 6 that with the signal processor 4 is interconnected. To power these electronic components is in the housing 2 also a battery 7 arranged, which is formed in the present embodiment concretely by an accumulator. To conduct the from the speaker 5 generated sound to the hearing of the hearing aid wearer is to the housing 2 a sound tube 8th connected, in the intended state of wear at the head 9 , specifically on the ear of the hearing aid wearer (cf. 2 ) with an earmold 10 is inserted into the ear canal of the hearing aid wearer.

The acceleration sensor 6 is designed for three-dimensional measurement, and has three perpendicular measuring axes x . y and z (S. 2 ) on. The acceleration sensor 6 is so in the case 2 of the hearing aid 1 arranged that in the intended state of wear on the head 9 and in the upright posture of the hearing aid wearer the measuring axis z at least approximately coincident with the vertical direction. The measuring axis x is radial to the head 9 , specifically directed radially outward. The measuring axis y is in this case tangential to the head 9 and oriented backwards - ie, in parallel opposite to a zero-degree viewing direction 12 in a basic orientation parallel to a sagittal plane, concretely parallel to the median plane 14 is aligned. The two measuring axes x and y are assigned to a transverse plane of the hearing aid wearer.

The signal processor 4 is set up by means of an acoustic classifier acting as an algorithm in the signal processor 4 implemented by means of the microphones 3 to record recorded sounds on a conversation situation (ie on a conversation of at least two people) and then adjust the signal processing accordingly. For example, this is an opening angle of a means of the two microphones 3 formed directional microphones adjusted so that all of the environment on the microphones 3 the relevant speech components are located within the opening range of the directional microphone. In order to be able to adapt the signal processing even more precisely in such a conversation situation, specifically to be able to set the aperture angle so that only the persons actually involved in the conversation (each representing a source location of a speech component) are within the aperture range of the directional microphone, the signal processor will 4 a below by means of 3 more detailed method performed.

In a first process step 20 are those of the accelerometer 6 determined measured values - those to samples of three measured values, of which in turn one each, one of the measuring axes x . y and z is assigned - as "raw" acceleration data A in a buffer memory (the one in the signal processor 4 integrated). The buffer memory is designed for the sliding intermediate storage of eight such samples, each of which is a data frame D and recorded over a 1.3 second window.

The raw acceleration data A contain, among other things information about an acceleration in straight, straight ahead (ie along the meridian plane 14 ) directed movement (eg walking and the like), over a radial to the head 9 directed acceleration (hereinafter referred to as "radial acceleration ar") and a tangent to the head 9 directed acceleration (hereinafter referred to as "tangential acceleration at"). The radial acceleration ar occurs due to the centrifugal force in a body and / or head rotation - regardless of the direction of rotation or rotation and thus independent of the associated axis of rotation - on. The tangential acceleration at also occurs during a rotation of the head 9 , specifically when "loosening" and "slowing down" the head 9 on. The following are the radial and tangential accelerations ar respectively. at with respect to a rotation of the head called "yaw" 9 around its vertical axis, which is approximately perpendicular to the transverse plane and as a rotation axis in this context also referred to as "yaw axis", used and described. Due to the above-described assignment of the measuring axes x and y to the transverse plane can be under certain conditions (specifically when a yaw motion is present) that of the measuring axis x assigned measured values of the radial acceleration ar and the measuring axis y assigned measured values according to the tangential acceleration at to assign. Also included are the raw acceleration data A a bias (or: "bias proportion"), which usually fluctuates over time due to temperature influences, as well as influences of different measurement sensitivity for the individual measuring axes x . y and z , In addition, the acceleration data included A the influence of gravitational acceleration G (also: gravity).

wobei u_x, u_y, u_z die jeweilige Achse angeben.In a subsequent process step 30 become the measuring axes x . y and z of the acceleration sensor 6 , concretely so the sensor coordinate system on the head 9 assigned coordinate system normalized. This is - specifically during an adaptation of the hearing aid 1 to the hearing aid wearer - during a phase in which the head 9 in straight posture straight out (view along the Meridianebene 14 ), the scalar and cross product between that from the acceleration data A determined gravitational vector and the global, one called "world coordinate system", universal coordinate system associated gravitational vector gw determined ,. The global gravitational vector gw is based on the assumption that gravitation shows "upward" and thus the shape gw = [0, 0, 1]. The scalar product gives a tilt angle between these vectors and the cross product a tilt axis for this tilt angle is effective, again. By means of the tilting axis and the tilting angle, a rotation operator, specifically a quaternion designated as a "normalization quaternion Qa", is constructed, which reproduces a combination of the sensor coordinate system with the world coordinate system and by means of which the sensor coordinate system and thus the raw acceleration data A be twisted ("normalized") to the world coordinate system. The normalization quaternion Qa is constructed on the basis of a so-called axis angle representation, namely that according to Euler every rotation can be represented by an angle around a fixed axis: $$Qa=\left(\begin{array}{c}\text{cos}\frac{\theta }{2}\\ u_x\text{sin}\frac{\theta }{2}\\ u_y\text{sin}\frac{\theta }{2}\\ u_z\text{sin}\frac{\theta }{2}\end{array}\right)$$

where u _x , u _y , u _z indicate the respective axis.

In addition, in an optional embodiment, a calibration, specifically an ellipsoid calibration is performed. In ellipsoidal calibration, it is assumed that the norm of the vector of gravitational acceleration G one yields, so that no explicit knowledge of the alignment of the measuring axes x . y and z of the acceleration sensor 6 is required. Other calibration methods that involve the actual alignment of the measuring axes x . y and z of the acceleration sensor 6 can be determined, but can be used equally. For example, for calibration for specific positioning of the hearing aid 1 and thus the measuring axes x . y and z of the acceleration sensor 6 or (even before mounting the acceleration sensor 6 in the case 2 ) only the acceleration sensor 6 the respective associated acceleration data A and uses it to determine calibration data for the bias and sensitivity effects. The specific positions are usually "set" by means of a handling device (also: "manipulator"). The respective measuring points assigned to the positions lie on a three-dimensional ellipsoid surface, which is usually deformed into an ellipsoid due to the different measuring sensitivities and assigned to the standard of the acceleration signal. This calibration allows the bias and sensitivity effects to be sufficiently accurately compensated during a "static" data frame (where no motion is detected).

Furthermore, in the process step 30 the acceleration data A a data frame D preprocessed. For this purpose, a noise reduction takes place by means of a median filter with a window length of three samples in order to filter out sensor noise and individual signal peaks.

In a further process step 40 become the (now preprocessed) acceleration data A within the respective data frame D analyzed and checked for the presence of a movement. For this purpose, a range of values in the data frame D contained acceleration data A Specifically, the distance between maximum and minimum values of the acceleration data is determined. In addition (or alternatively in a different embodiment), a variance of the acceleration data also becomes A of each data frame D determined. A comparison of thresholds is used to determine whether the range of values and / or the variance is sufficiently different from a corresponding value range or a corresponding variance value of a measurement noise that contains the acceleration signal or the acceleration data A dominated in the absence of movement of the hearing aid wearer, take off.

In the event that a movement of the hearing aid wearer is detected, the acceleration data becomes A in a process step 50 to the presence of a yawing motion of the head 9 examined. With a yawing motion of the head 9 (ie, a rotation about an axis perpendicular to the transverse plane, usually around the axis formed by the spinal column) is shown in FIG 4 schematically over time t shown course of the (normalized to the world coordinate system) measuring axes x . y and z associated radial and tangential accelerations ar . at , which represent the centrifugal force and the tangential force during the yawing motion, as well as the gravitational acceleration G , which is positive here. When yawing to the left the tangential acceleration passes through at a first extremum indicative of the beginning of the yaw motion, and then a second extremum of opposite sign indicating the deceleration of the head 9 indicates the end of the yaw movement. Parallel shows the radial acceleration ar for the centrifugal force also a single extremum. The period of yawing is within about 1.2 seconds. Already this course of the radial and tangential accelerations ar and at is used in an optional embodiment as a criterion for the presence of the yawing motion.

In 5 are the ones for the accelerations ar . at and G recorded measured values of the measuring axes x . y and z three-dimensional against each other. As can be seen, give the radial and the tangential acceleration ar respectively. at an ellipse-like curve around the acceleration due to gravity G in the measuring axis z is moved. In the process step 50 becomes a curve, concretely an ellipse on the mutually plotted course (the acceleration curve) of the radial and tangential accelerations ar respectively. at fit and used the form as a criterion for the presence of the yaw. If the shape of the fitted curve is sufficiently close, concretely with a small residual error, approximated to an ellipse, the probability of the existence of a yawing motion is high. In fact, other movements regularly express themselves in an irregular - usually zig-zag-like - course of the radial and tangential accelerations presented to one another ar and at , The area, specifically the surface area of the fitted ellipse, reflects the strength of the yawing motion, ie how fast the head is 9 is turned.

Subsequently, for the fitted ellipse, the normal vector, specifically the normal vector standing in the center of gravity of the ellipse surface, is determined. This corresponds to the axis of rotation or "yaw axis 52 "Of greed. As in 5 can be seen is the yaw axis 52 opposite the measuring axis z tilted, and thus also one described by the ellipse surface and by the radial and tangential accelerations ar and at spanned movement plane 54 , The tilt of the plane of motion 54 opposite one of the measuring axes x and y clamped measuring level 56 is determined - for example by means of the scalar and cross product of the yaw axis 52 with the measuring axis z , specifically their axis vector. By means of this tilting information, a "correction quaternion" for "correct rotation" of the plane of motion is used as a rotation operator for correcting the tilting 54 parallel to the measuring level 56 created, by means of this the acceleration data A parallel to the measuring axes x . y , and z aligned (rotated) and then the fitted ellipse on the measurement plane 56 projected. This allows the acceleration data A low-loss on the two-dimensional measuring plane 56 be imaged.

In the case where the presence of the yawing motion has been detected, the tangential acceleration becomes at integrated twice and thus determined the distance covered in the yaw motion. By the radius of the head 9 divided results in the current angular position of the head 9 , In addition, while the yaw angle associated with the current angular position G determined. From the angular position or the yaw angle G and the yaw axis 52 is constructed a quaternion called "motion quaternion Qb". This in turn represents a rotation operator that represents the movement of the head.

auf.In a subsequent process step 60 is using the in motion quaternion Qb contained motion information of a previous orientation of the head 9 associated orientation rotation operator, specifically an "orientation quaternion Qn-1" to a current orientation quaternion Qn "Updated", ie renewed by the previous orientation quaternion Qn-1 with the motion quaternion Qb is multiplied. As a "first" orientation quaternion Qn is a unit quaternion for initialization Qe assumed, ie assumed as fictitious "starting value". This has the shape $$Qe=\left(\begin{array}{c}1\\ 0\\ 0\\ 0\end{array}\right)$$

on.

Will in the process step 40 no movement detected takes place in a further method step 70 for the corresponding "static" data frame D a correction of the yaw angle G , and a pitch and a roll angle of the head 9 , First, this is the influence of gravitational acceleration G from the acceleration data A away. To measurement noise in the acceleration data A to decrease then becomes an average ames determined for this. Subsequently, the mean value ames Specifically, the vector assigned to it by means of the previous data frames D determined orientation, specifically by means of the above-described previous orientation quaternion Qn-1 rotates. By a residual deviation of the resulting rotated mean vector aw from the target orientation, given by the global gravitational vector gw , the scalar product and the cross product between the resulting mean vector will determine aw and the global gravitational vector gw determined. The scalar product results in a correction angle θk and the cross product the correction axis uw to make the one rotation with the correction angle θk must be done to the resulting mean vector aw with the global gravitational vector gw to bring to cover (cf. 6 ). This then becomes a deviation quaternion qd1 constructed.

In addition, the yaw angle G during a static data frame D comparatively slowly in the direction of the zero-degree viewing direction 12 drifted. For this purpose, in the process step 70 based on the global gravitational vector gw a greedy quaternion Qd2 determined. That is, this greedy quaternion Qd2 contains only the current information of the pitch and roll angles, while the yaw angle G to zero, that is, to the zero-degree viewing direction 12 is set. The zero-degree viewing direction 12 does not have to (or no longer) coincide with the sagittal plane (ie parallel to the median plane 14 be aligned. Specifically, this greedy quaternion Qd2 analogous to the deviation quaternion described above qd1 determined. Subsequently, between the greedy quaternion Qd2 and the deviation quaternion qd1 interpolated spherically, so that the longer no movement is detected, the current value of the yaw angle G zero degree is approximated. In other words, the zero-degree viewing direction 12 from the alignment parallel to the median plane 14 to the current angular position of the head 9 postponed. This interpolation becomes a new deviation quaternion qd educated. As a step size in this interpolation are - depending on the desired duration of the drift of the zero-degree viewing direction 12 to the current angular position of the head 9 values between, for example, 0.5 percent (slow drift) or 2 percent (relatively fast drift).

In the process step 60 is then analogous to the previously described update of the previous orientation quaternion Qn-1 to the current orientation quaternion Qn the new deviation quaternion qd (instead of the motion quaternion Qb ). Subsequently, a return to the method step takes place for the next data frame 40 ,

In a non-illustrated method step is based on the current yaw angle G creates a line of sight probability distribution (s. 7 ) for each current yaw angle G a Gaussian normal distribution with a standard deviation of z. B. 10 degrees is set. For every yaw angle G along which the hearing device wearer has aligned his gaze, the line of sight probability distribution thus contains a "peak", so that it is possible to read where the hearing device wearer has looked, for example, in the last minute. To a "forgetting" or "fading" comparatively "old" (eg older than 1 to 3 minutes) from the head 9 assumed yaw angle G allow, in an optional embodiment, before each new entry in the line of sight probability distribution, the previous probability values are uniformly reduced. The new normal distribution is entered accordingly weighted, so that all angular positions cumulatively one. By means of the line of sight probability distribution it can be estimated whether the hearing aid wearer is engaged in a conversation with several people, then he would repeatedly glance between individual yaw angles G have changed, participated or whether he "only" looks in one direction. This information is provided by the signal processor 4 for setting a directivity of the microphones 3 used.

The object of the invention is not limited to the embodiments described above. Rather, other embodiments of the invention may be derived by those skilled in the art from the foregoing description. In particular, the individual features of the invention described with reference to the various exemplary embodiments and their design variants can also be combined with one another in a different manner.

LIST OF REFERENCE NUMBERS

1

hearing Aid

2

casing

3

microphone

4

signal processor

5

speaker

6

accelerometer

7

battery

8th

sound tube

9

head

10

earpiece

12

Zero degree line of sight

14

median plane

20

step

30

step

40

step

50

step

52

yaw axis

54

movement plane

56

measuring plane

60

step

70

step

A

acceleration data

ames

Average

ar

Radial acceleration

at

Tangential acceleration

aw

resulting mean vector

D

data frame

G

yaw

G

acceleration of gravity

gw

global gravity vector

Qa

Normalization quaternion

Qb

Motion quaternion

Qd, Qd1

Deviation quaternion

Qd2

greedy quaternion

Qe

unit quaternion

Qn, Qn-1

Orientation quaternion

t

Time

θk

Correction angle

uw

Correction shaft

x, y, z

measuring axis

Claims (26)

Method for operating a hearing device (1), which has an acceleration sensor (6) which is positioned in the intended wearing state on the head (9) of a hearing aid wearer and which is set up for measurement in three measuring axes (x, y, z) perpendicular to one another, wherein according to the method is concluded on the basis of transmitted by an acceleration signal acceleration data (A) of the acceleration sensor (6) on a movement of the hearing aid wearer, from the acceleration data (A) a movement plane (54) of the movement of the hearing device wearer is derived, - From the acceleration data (A) a movement axis (52) and a movement direction of the movement are determined, and - Based on the movement plane (54), the movement axis (52) and the direction of movement is closed to a presence of a rotational movement of the head (9). Method according to Claim 1 in which the presence of a yawing motion is used as the presence of the rotational movement. Method according to Claim 1 or 2 in which the acceleration data (A) are analyzed in blocks in successive data frames (D). Method according to one of Claims 1 to 3 , Wherein a value range and / or a variance of the acceleration signal are used as a feature for the presence of the movement of the hearing aid wearer. Method according to one of Claims 1 to 4 , wherein in the proper wearing state of the hearing aid (1) from the acceleration data (A) for a rotation of the measuring axes (x, y, z) of the acceleration sensor (6) relative to a transverse plane of the body of the hearing aid wearer and / or a horizontal plane indicative normalization rotation operator (Qa) is determined. Method according to Claim 5 in which the rotation of the measuring axes (x, y, z) is determined on the basis of the cross product and the scalar product between the gravitational vector determined from the acceleration data (A) and a global gravitational vector (gw). Method according to one of Claims 1 to 6 in which the acceleration data (A) are smoothed, in particular by means of a median filter. Method according to one of Claims 1 to 7 , wherein for determining the rotational movement, in particular a yaw movement of the head (9) from the acceleration data (A) directed for the rotational movement radially to the head (9) of the hearing aid wearer acceleration (ar) and a tangentially directed acceleration (at) are derived. Method according to Claim 8 wherein the plane of motion (54) is constructed from the radially directed acceleration (ar) and the tangentially directed acceleration (at), and wherein the plane of motion (54) is rotated to be parallel to one of the two for determining the radially directed one Acceleration (ar) and the tangentially directed acceleration (at) used measuring axes (x, y, z) clamped measuring plane (56) is aligned. Method according to Claim 9 , wherein for the rotation of the plane of motion (54) the tilting of the plane of movement (54) relative to the measuring plane (56) is determined by an ellipse in a mutually applied course of the radially directed acceleration (ar) and the tangentially directed acceleration (at) fitted is assumed, a normal vector of the ellipse surface as the movement axis (52) and for the normal vector, a tilt angle with respect to the measurement plane (56) is determined. Method according to Claim 10 in which a motion rotation operator (Qb) for the current movement head (9) of the hearing aid wearer is constructed by means of the movement plane (54) by determining a rotation angle (G) of the rotational movement of the head (9) on the basis of the tangentially directed acceleration (at) becomes. Method according to Claim 10 or 11 , wherein based on the tangentially directed acceleration (at) only when the presence of the rotational movement is detected, the rotation angle (G) of the rotational movement of the head (9) is determined. Method according to one of Claims 9 to 12 in that as a feature for the presence of the rotational movement of the head (9), a surface of the ellipse is determined and closed depending on the characteristic of the feature on the presence of the rotational movement. Method according to one of Claims 9 to 13 in which, as a feature for the presence of the rotational movement of the head (9) of the hearing aid wearer, an associated intensity and / or an intensity of a tangential velocity is derived from the tangentially directed acceleration (at). Method according to one of Claims 1 to 14 , Wherein, in the event that no movement of the hearing aid wearer is detected, a correction of a pitch angle and / or a roll angle of the head (9), whose associated axes of rotation are each at least roughly in the transverse plane is performed. Method according to Claim 15 in which the acceleration data (A) of the current data frame (D) is rotated by means of the alignment information of the preceding data frame (D), and wherein for the thus rotated acceleration data (A), respectively, by means of the cross product and the dot product, a residual deviation of the orientation compared to previous data frame (D) is determined. Method according to Claim 16 in which a deviation rotation operator (Qd1) is constructed by the residual deviation. Method according to one of Claims 15 to 17 wherein, in the event that no movement of the hearing aid wearer is detected, a correction of the yaw movement associated yaw angle (G) is performed by a main viewing direction (12) of the hearing aid wearer away from the median plane (14) of the body of the hearing aid wearer to a current viewing direction adjusted. Method according to Claim 16 and 18 wherein a yaw-free rotation operator (Qd2) in which the yaw angle (G) is set to zero degrees is determined on the basis of the global gravitational vector (gw), and wherein between the yaw-free rotation operator (Qd2) and the deviation rotation operator (Qd1) is spherically interpolated to form a new deviation rotation operator (Qd). Method according to one of Claims 11 to 19 in which, for the current orientation of the head (9), a new orientation rotation operator (Qn) is determined by using a preceding orientation rotation operator (Qn-1) with the motion rotation operator (Qb) or with the deviation rotation operator (Qd, Qd1) is charged. Method according to one of Claims 5 to 20 , wherein as a rotation operator in each case a corresponding quaternion (Qa, Qb, Qd, Qd1, Qd2, Qn, Qn-1) is determined. Method according to Claim 20 or 21 wherein, based on the orientation rotation operator (Qn), a view-direction likelihood distribution is created. Method according to Claim 22 in which, starting from the main viewing direction, a Gaussian normal distribution with a predetermined standard deviation is assumed, and wherein for a current yaw position a new normal distribution with a predetermined standard deviation is set. Method according to Claim 10 and 23 , wherein the standard deviation is set variably as a function of a residual error of the determination of the movement plane (54). Method according to one of Claims 1 to 24 in which the information about the rotational movement of the head (9) of the hearing device wearer, in particular the line of sight probability distribution, is used to adapt a signal processing algorithm for a group talk situation. Hearing aid (1) with an acceleration sensor (6) which is positioned in the intended wearing state on the head (9) of a hearing aid wearer and which is set up for measurement in three measuring axes (x, y, z) perpendicular to one another and with a processor (4) responsible for carrying out the method according to one of Claims 1 to 25 is set up.

Images